Selflearning controller of active magnetic bearing

Selflearning controller of active magnetic bearing

Paper in proceedings (conference paper)

The active magnetic bearing control through self learning controller is described in this contribution. controller's coefficient (parameter) values come from actions of Continuous Action Reinforcement Learning Automatas (CARLAs) whic continuously update the controller's coefficients according to behavior of the active magnetic bearing. The goal of this on-line training is formulated as achievement of minimum mean square of control error. It is shown that CARLA method is capable of learning better parameters than standard method of optimal control design called LQ (linear quadratic) design. Described concept of control is proved by control of the active magnetic bearing.

The active magnetic bearing control through self learning controller is described in this contribution. controller's coefficient (parameter) values come from actions of Continuous Action Reinforcement Learning Automatas (CARLAs) whic continuously update the controller's coefficients according to behavior of the active magnetic bearing. The goal of this on-line training is formulated as achievement of minimum mean square of control error. It is shown that CARLA method is capable of learning better parameters than standard method of optimal control design called LQ (linear quadratic) design. Described concept of control is proved by control of the active magnetic bearing.

Active Magnetic Bearing, Continuous Action Reinforcement Learning Automata

Active Magnetic Bearing, Continuous Action Reinforcement Learning Automata

TUREK, M.; BŘEZINA, T.; PULCHART, J.

15.05.2006

Book of extended abstracts

392

393

2